Artificial Intelligence by Michael Wooldridge

Author:Michael Wooldridge
Language: eng
Format: epub
ISBN: 9781405934183
Publisher: Penguin Books Ltd
Published: 2018-02-13T16:00:00+00:00

Coping with uncertainty

The Turing test established human behaviour as the goal of AI. But humans are often poor decision-makers – why should we want to build systems that make bad decisions? The goal of AI began to shift from making human decisions to making rational decisions.

A key aspect of rational decision-making relates to reasoning with uncertain information. People are bad at this. Consider:

A deadly new flu virus infects one in every thousand people. A test for the flu is developed, which is 99 per cent accurate. On a whim you take the test, and it comes out positive.

Most people would be very worried, but paradoxically there is only about a one in ten chance that you have the flu. Why? Because the prior probability that you had the flu was one in a thousand – far more people don’t have flu than do, so there will be many more false positives than true positives.

The basic mathematics behind this reasoning was developed by the Reverend Thomas Bayes in the eighteenth century, but much work was needed to make Bayesian reasoning usable in AI, because AI systems often have to deal with lots of structured evidence.

Many automated translation systems use Bayesian reasoning. They compute the likeliest translation of a word, given the words that have appeared previously. Prior probabilities are computed by examining many translated texts. These translation tools are very useful for routine tasks, but in no sense do they understand the text they translate.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.